Klavis AI

Klavis AI supplies coding and tool-use training data and environments for frontier AI agents. Its current company offering focuses on long-horizon tasks and verifiable rewards for post-training. The separate MCP documentation and FAQ describe integration products; their free and paid tiers should not be applied to training-data contracts.

✨ Key Features

• Coding data — Tasks cover editing code, writing tests and debugging, with programmatic verification, deterministic tests and granular rewards.
• Packaged environments — Docker environments support reinforcement learning and supervised fine-tuning workflows.
• Agentic tool-use data — The company lists 600+ real tools and SaaS apps, production MCP servers, state-changing workflows and logically consistent task state.
• Reward methods — Tool-use tasks can use deterministic, rubric-based and LLM-judge rewards.
• Separate MCP surface — API documentation still lists Strata creation, server connections and authentication controls. This does not establish that every earlier integration or commercial offer remains available unchanged.

💳 Pricing

Training-data and environment engagements use a contact entry; fixed public prices, included volumes and a free training-data tier are not established. The MCP FAQ separately describes a basic free option, Pro and Enterprise tiers, without current fixed amounts. Those descriptions apply to the MCP product surface, not the current data offering.

⚖️ Products at Different Layers

🎯 Use Cases

• Training coding agents on longer code, test and debugging tasks.
• Developing tool-use agents with realistic service state and reward signals.
• Exploring the separately documented MCP server APIs for an integration project.

MCP API documentation is a separate developer reference. Confirm the relevant product, access and commercial terms with Klavis.